Developing Seismic Fragility Curves Using ANN Based Probabilistic Seismic Demand Models Derived from Structural Design Parameters

Assessing the fragility and damage state of multiple buildings in an urban setting remains a challenging task requiring considerable time and cost. This study proposes deriving seismic fragility curves using an Artificial Neural Network (ANN) based Probabilistic Seismic Demand Model (PSDM) to overcome these challenges. Seismic fragility curves were developed using the ANN based PSDM to derive the regression function of interstory drift and spectral acceleration. The methodology involves conducting nonlinear dynamic analysis for 540 steel moment frames(SMFs) using 240 seismic records to construct a PSDM for each SMF. The ANN-based PSDM was developed using nine design variables (number of stories, number of bays, bay width, first-story height, floor dead load, roof dead load, and first to third natural periods of SMFs) as input and the regression function of interstory drift and spectral acceleration as output. Fragility curves for SMFs were derived using the ANN-based PSDM. ANN-based PSDM exhibited an accuracy of R-value 0.96 for the training database. The developed ANN-based PSDM is validated and compared with the results obtained from the general method using nonlinear dynamic analysis. The results show that the ANN-based PSDM accurately predicts damage states and the fragility curves derived using this method are consistent with those obtained from nonlinear dynamic analysis. The proposed methodology offers a time-efficient and reliable approach for assessing the fragility of SMFs without the need for detailed structural modeling and time-consuming nonlinear analysis.